8 #include <sys/resource.h>
17 #include "compiler/compiler.h"
18 #include "thread_options.h"
21 #include "arch/arch.h"
28 #include "ioengines.h"
35 #include "oslib/getopt.h"
37 #include "lib/rbtree.h"
38 #include "lib/num2str.h"
44 #include "io_u_queue.h"
45 #include "workqueue.h"
46 #include "steadystate.h"
48 #ifdef CONFIG_SOLARISAIO
49 #include <sys/asynch.h>
53 #include <linux/mempolicy.h>
57 * "local" is pseudo-policy
59 #define MPOL_LOCAL MPOL_MAX
67 * offset generator types
75 __TD_F_VER_BACKLOG = 0,
78 __TD_F_REFILL_BUFFERS,
79 __TD_F_SCRAMBLE_BUFFERS,
91 __TD_F_LAST, /* not a real bit, keep last */
95 TD_F_VER_BACKLOG = 1U << __TD_F_VER_BACKLOG,
96 TD_F_TRIM_BACKLOG = 1U << __TD_F_TRIM_BACKLOG,
97 TD_F_READ_IOLOG = 1U << __TD_F_READ_IOLOG,
98 TD_F_REFILL_BUFFERS = 1U << __TD_F_REFILL_BUFFERS,
99 TD_F_SCRAMBLE_BUFFERS = 1U << __TD_F_SCRAMBLE_BUFFERS,
100 TD_F_VER_NONE = 1U << __TD_F_VER_NONE,
101 TD_F_PROFILE_OPS = 1U << __TD_F_PROFILE_OPS,
102 TD_F_COMPRESS = 1U << __TD_F_COMPRESS,
103 TD_F_COMPRESS_LOG = 1U << __TD_F_COMPRESS_LOG,
104 TD_F_VSTATE_SAVED = 1U << __TD_F_VSTATE_SAVED,
105 TD_F_NEED_LOCK = 1U << __TD_F_NEED_LOCK,
106 TD_F_CHILD = 1U << __TD_F_CHILD,
107 TD_F_NO_PROGRESS = 1U << __TD_F_NO_PROGRESS,
108 TD_F_REGROW_LOGS = 1U << __TD_F_REGROW_LOGS,
109 TD_F_MMAP_KEEP = 1U << __TD_F_MMAP_KEEP,
110 TD_F_DIRS_CREATED = 1U << __TD_F_DIRS_CREATED,
121 FIO_RAND_FILE_SIZE_OFF,
124 FIO_RAND_SEQ_RAND_READ_OFF,
125 FIO_RAND_SEQ_RAND_WRITE_OFF,
126 FIO_RAND_SEQ_RAND_TRIM_OFF,
127 FIO_RAND_START_DELAY,
129 FIO_RAND_POISSON_OFF,
131 FIO_RAND_POISSON2_OFF,
132 FIO_RAND_POISSON3_OFF,
140 RATE_PROCESS_LINEAR = 0,
141 RATE_PROCESS_POISSON = 1,
152 * Per-thread/process specific data. Only used for the network client
155 void sk_out_assign(struct sk_out *);
156 void sk_out_drop(void);
158 struct zone_split_index {
160 uint8_t size_perc_prev;
166 * This describes a single thread/process executing a fio job.
169 struct flist_head opt_list;
171 struct thread_options o;
174 unsigned int thread_number;
175 unsigned int subjob_number;
176 unsigned int groupid;
177 struct thread_stat ts __attribute__ ((aligned(8)));
181 struct io_log *slat_log;
182 struct io_log *clat_log;
183 struct io_log *clat_hist_log;
184 struct io_log *lat_log;
185 struct io_log *bw_log;
186 struct io_log *iops_log;
188 struct workqueue log_compress_wq;
190 struct thread_data *parent;
192 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
193 struct timespec bw_sample_time;
195 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
196 struct timespec iops_sample_time;
198 volatile int update_rusage;
199 struct fio_mutex *rusage_sem;
200 struct rusage ru_start;
201 struct rusage ru_end;
203 struct fio_file **files;
204 unsigned char *file_locks;
205 unsigned int files_size;
206 unsigned int files_index;
207 unsigned int nr_open_files;
208 unsigned int nr_done_files;
209 unsigned int nr_normal_files;
211 unsigned int next_file;
212 struct frand_state next_file_state;
215 struct zipf_state next_file_zipf;
216 struct gauss_state next_file_gauss;
229 size_t orig_buffer_size;
230 volatile int terminate;
231 volatile int runstate;
232 unsigned int last_was_sync;
233 enum fio_ddir last_ddir;
240 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
242 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
243 struct frand_state verify_state;
244 struct frand_state trim_state;
245 struct frand_state delay_state;
247 struct frand_state buf_state;
248 struct frand_state buf_state_prev;
249 struct frand_state dedupe_state;
250 struct frand_state zone_state;
252 struct zone_split_index **zone_state_index;
254 unsigned int verify_batch;
255 unsigned int trim_batch;
257 struct thread_io_list *vstate;
262 * IO engine hooks, contains everything needed to submit an io_u
263 * to any of the available IO engines.
265 struct ioengine_ops *io_ops;
269 * IO engine private data and dlhandle.
272 void *io_ops_dlhandle;
275 * Queue depth of io_u's that fio MIGHT do
277 unsigned int cur_depth;
280 * io_u's about to be committed
282 unsigned int io_u_queued;
285 * io_u's submitted but not completed yet
287 unsigned int io_u_in_flight;
290 * List of free and busy io_u's
292 struct io_u_ring io_u_requeues;
293 struct io_u_queue io_u_freelist;
294 struct io_u_queue io_u_all;
295 pthread_mutex_t io_u_lock;
296 pthread_cond_t free_cond;
299 * async verify offload
301 struct flist_head verify_list;
302 pthread_t *verify_threads;
303 unsigned int nr_verify_threads;
304 pthread_cond_t verify_cond;
305 int verify_thread_exit;
310 uint64_t rate_bps[DDIR_RWDIR_CNT];
311 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
312 unsigned long rate_bytes[DDIR_RWDIR_CNT];
313 unsigned long rate_blocks[DDIR_RWDIR_CNT];
314 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
315 struct timespec lastrate[DDIR_RWDIR_CNT];
316 int64_t last_usec[DDIR_RWDIR_CNT];
317 struct frand_state poisson_state[DDIR_RWDIR_CNT];
320 * Enforced rate submission/completion workqueue
322 struct workqueue io_wq;
324 uint64_t total_io_size;
325 uint64_t fill_device_size;
330 uint64_t io_issues[DDIR_RWDIR_CNT];
331 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
337 uint64_t io_blocks[DDIR_RWDIR_CNT];
338 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
339 uint64_t io_bytes[DDIR_RWDIR_CNT];
340 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
341 uint64_t io_skip_bytes;
343 struct fio_mutex *mutex;
344 uint64_t bytes_done[DDIR_RWDIR_CNT];
347 * State for random io, a bitmap of blocks done vs not done
349 struct frand_state random_state;
351 struct timespec start; /* start of this loop */
352 struct timespec epoch; /* time job was started */
353 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
354 struct timespec last_issue;
356 struct timespec ts_cache;
357 struct timespec terminate_time;
358 unsigned int ts_cache_nr;
359 unsigned int ts_cache_mask;
363 * Time since last latency_window was started
365 struct timespec latency_ts;
366 unsigned int latency_qd;
367 unsigned int latency_qd_high;
368 unsigned int latency_qd_low;
369 unsigned int latency_failed;
370 uint64_t latency_ios;
374 * read/write mixed workload state
376 struct frand_state rwmix_state;
377 unsigned long rwmix_issues;
378 enum fio_ddir rwmix_ddir;
379 unsigned int ddir_seq_nr;
382 * rand/seq mixed workload state
384 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
387 * IO history logs for verification. We use a tree for sorting,
388 * if we are overwriting. Otherwise just use a fifo.
390 struct rb_root io_hist_tree;
391 struct flist_head io_hist_list;
392 unsigned long io_hist_len;
397 struct flist_head io_log_list;
400 * For tracking/handling discards
402 struct flist_head trim_list;
403 unsigned long trim_entries;
405 struct flist_head next_rand_list;
408 * for fileservice, how often to switch to a new file
410 unsigned int file_service_nr;
411 unsigned int file_service_left;
412 struct fio_file *file_service_file;
414 unsigned int sync_file_range_nr;
417 * For generating file sizes
419 struct frand_state file_size_state;
424 unsigned int total_err_count;
427 struct fio_flow *flow;
430 * Can be overloaded by profiles
432 struct prof_io_ops prof_io_ops;
437 struct steadystate_data ss;
439 char verror[FIO_VERROR_SIZE];
443 * for GPU memory management
449 CUdeviceptr dev_mem_ptr;
455 * when should interactive ETA output be generated
463 #define __td_verror(td, err, msg, func) \
465 unsigned int ____e = (err); \
468 (td)->error = ____e; \
469 if (!(td)->first_error) \
470 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
474 #define td_clear_error(td) do { \
477 (td)->parent->error = 0; \
480 #define td_verror(td, err, func) do { \
481 __td_verror((td), (err), strerror((err)), (func)); \
483 __td_verror((td)->parent, (err), strerror((err)), (func)); \
486 #define td_vmsg(td, err, msg, func) do { \
487 __td_verror((td), (err), (msg), (func)); \
489 __td_verror((td)->parent, (err), (msg), (func)); \
492 #define __fio_stringify_1(x) #x
493 #define __fio_stringify(x) __fio_stringify_1(x)
495 extern int exitall_on_terminate;
496 extern unsigned int thread_number;
497 extern unsigned int stat_number;
500 extern int output_format;
501 extern int append_terse_output;
502 extern int temp_stall_ts;
503 extern uintptr_t page_mask, page_size;
504 extern int read_only;
505 extern int eta_print;
506 extern int eta_new_line;
507 extern unsigned long done_secs;
508 extern int fio_gtod_offload;
509 extern int fio_gtod_cpu;
510 extern enum fio_cs fio_clock_source;
511 extern int fio_clock_source_set;
512 extern int warnings_fatal;
513 extern int terse_version;
514 extern int is_backend;
515 extern int nr_clients;
516 extern int log_syslog;
517 extern int status_interval;
518 extern const char fio_version_string[];
519 extern char *trigger_file;
520 extern char *trigger_cmd;
521 extern char *trigger_remote_cmd;
522 extern long long trigger_timeout;
523 extern char *aux_path;
525 extern struct thread_data *threads;
527 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
529 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
532 #define REAL_MAX_JOBS 4096
534 static inline int should_fsync(struct thread_data *td)
536 if (td->last_was_sync)
538 if (td_write(td) || td->o.override_sync)
545 * Init/option functions
547 extern int __must_check fio_init_options(void);
548 extern int __must_check parse_options(int, char **);
549 extern int parse_jobs_ini(char *, int, int, int);
550 extern int parse_cmd_line(int, char **, int);
551 extern int fio_backend(struct sk_out *);
552 extern void reset_fio_state(void);
553 extern void clear_io_state(struct thread_data *, int);
554 extern int fio_options_parse(struct thread_data *, char **, int);
555 extern void fio_keywords_init(void);
556 extern void fio_keywords_exit(void);
557 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
558 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
559 extern void fio_fill_default_options(struct thread_data *);
560 extern int fio_show_option_help(const char *);
561 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
562 extern void fio_options_dup_and_init(struct option *);
563 extern void fio_options_mem_dupe(struct thread_data *);
564 extern void td_fill_rand_seeds(struct thread_data *);
565 extern void td_fill_verify_state_seed(struct thread_data *);
566 extern void add_job_opts(const char **, int);
567 extern int ioengine_load(struct thread_data *);
568 extern bool parse_dryrun(void);
569 extern int fio_running_or_pending_io_threads(void);
570 extern int fio_set_fd_nonblocking(int, const char *);
571 extern void sig_show_status(int sig);
572 extern struct thread_data *get_global_options(void);
574 extern uintptr_t page_mask;
575 extern uintptr_t page_size;
576 extern int initialize_fio(char *envp[]);
577 extern void deinitialize_fio(void);
579 #define FIO_GETOPT_JOB 0x89000000
580 #define FIO_GETOPT_IOENGINE 0x98000000
581 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
586 extern void print_thread_status(void);
587 extern void print_status_init(int);
588 extern char *fio_uint_to_kmg(unsigned int val);
591 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
592 * will never back again. It may cycle between running/verififying/fsyncing.
593 * Once the thread reaches TD_EXITED, it is just waiting for the core to
613 #define TD_ENG_FLAG_SHIFT 16
614 #define TD_ENG_FLAG_MASK ((1U << 16) - 1)
616 static inline void td_set_ioengine_flags(struct thread_data *td)
618 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
619 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
622 static inline bool td_ioengine_flagged(struct thread_data *td,
623 enum fio_ioengine_flags flags)
625 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
628 extern void td_set_runstate(struct thread_data *, int);
629 extern int td_bump_runstate(struct thread_data *, int);
630 extern void td_restore_runstate(struct thread_data *, int);
631 extern const char *runstate_to_name(int runstate);
634 * Allow 60 seconds for a job to quit on its own, otherwise reap with
637 #define FIO_REAP_TIMEOUT 300
639 #define TERMINATE_ALL (-1U)
640 extern void fio_terminate_threads(unsigned int);
641 extern void fio_mark_td_terminate(struct thread_data *);
646 extern int __must_check fio_pin_memory(struct thread_data *);
647 extern void fio_unpin_memory(struct thread_data *);
648 extern int __must_check allocate_io_mem(struct thread_data *);
649 extern void free_io_mem(struct thread_data *);
650 extern void free_threads_shm(void);
653 #define PTR_ALIGN(ptr, mask) \
654 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
658 * Reset stats after ramp time completes
660 extern void reset_all_stats(struct thread_data *);
662 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
663 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
664 struct timespec *comp_time);
667 * Latency target helpers
669 extern void lat_target_check(struct thread_data *);
670 extern void lat_target_init(struct thread_data *);
671 extern void lat_target_reset(struct thread_data *);
674 * Iterates all threads/processes within all the defined jobs
676 #define for_each_td(td, i) \
677 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
678 #define for_each_file(td, f, i) \
679 if ((td)->files_index) \
680 for ((i) = 0, (f) = (td)->files[0]; \
681 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
684 #define fio_assert(td, cond) do { \
687 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
688 td_set_runstate((td), TD_EXITED); \
689 (td)->error = EFAULT; \
694 static inline bool fio_fill_issue_time(struct thread_data *td)
696 if (td->o.read_iolog_file ||
697 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
703 static inline bool __should_check_rate(struct thread_data *td,
706 struct thread_options *o = &td->o;
709 * If some rate setting was given, we need to check it
711 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
712 o->rate_iops_min[ddir])
718 static inline bool should_check_rate(struct thread_data *td)
720 if (td->bytes_done[DDIR_READ] && __should_check_rate(td, DDIR_READ))
722 if (td->bytes_done[DDIR_WRITE] && __should_check_rate(td, DDIR_WRITE))
724 if (td->bytes_done[DDIR_TRIM] && __should_check_rate(td, DDIR_TRIM))
730 static inline unsigned int td_max_bs(struct thread_data *td)
734 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
735 return max(td->o.max_bs[DDIR_TRIM], max_bs);
738 static inline unsigned int td_min_bs(struct thread_data *td)
742 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
743 return min(td->o.min_bs[DDIR_TRIM], min_bs);
746 static inline bool td_async_processing(struct thread_data *td)
748 return (td->flags & TD_F_NEED_LOCK) != 0;
752 * We currently only need to do locking if we have verifier threads
753 * accessing our internal structures too
755 static inline void td_io_u_lock(struct thread_data *td)
757 if (td_async_processing(td))
758 pthread_mutex_lock(&td->io_u_lock);
761 static inline void td_io_u_unlock(struct thread_data *td)
763 if (td_async_processing(td))
764 pthread_mutex_unlock(&td->io_u_lock);
767 static inline void td_io_u_free_notify(struct thread_data *td)
769 if (td_async_processing(td))
770 pthread_cond_signal(&td->free_cond);
773 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
776 if (!td_async_processing(td))
779 __sync_fetch_and_and(flags, ~value);
782 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
785 if (!td_async_processing(td))
788 __sync_fetch_and_or(flags, value);
791 extern const char *fio_get_arch_string(int);
792 extern const char *fio_get_os_string(int);
795 #define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
796 #define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
800 __FIO_OUTPUT_TERSE = 0,
801 __FIO_OUTPUT_JSON = 1,
802 __FIO_OUTPUT_NORMAL = 2,
803 __FIO_OUTPUT_JSON_PLUS = 3,
806 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
807 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
808 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
809 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
813 FIO_RAND_DIST_RANDOM = 0,
815 FIO_RAND_DIST_PARETO,
818 FIO_RAND_DIST_ZONED_ABS,
821 #define FIO_DEF_ZIPF 1.1
822 #define FIO_DEF_PARETO 0.2
825 FIO_RAND_GEN_TAUSWORTHE = 0,
827 FIO_RAND_GEN_TAUSWORTHE64,
835 extern void exec_trigger(const char *);
836 extern void check_trigger_file(void);